1 //===-- SymbolizableObjectFile.cpp ----------------------------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // Implementation of SymbolizableObjectFile class.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "SymbolizableObjectFile.h"
15 #include "llvm/Object/SymbolSize.h"
16 #include "llvm/Support/DataExtractor.h"
17 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
18 
19 namespace llvm {
20 namespace symbolize {
21 
22 using namespace object;
23 
24 static DILineInfoSpecifier
25 getDILineInfoSpecifier(FunctionNameKind FNKind) {
26   return DILineInfoSpecifier(
27       DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, FNKind);
28 }
29 
30 ErrorOr<std::unique_ptr<SymbolizableObjectFile>>
31 SymbolizableObjectFile::create(object::ObjectFile *Obj,
32                                std::unique_ptr<DIContext> DICtx) {
33   std::unique_ptr<SymbolizableObjectFile> res(
34       new SymbolizableObjectFile(Obj, std::move(DICtx)));
35   std::unique_ptr<DataExtractor> OpdExtractor;
36   uint64_t OpdAddress = 0;
37   // Find the .opd (function descriptor) section if any, for big-endian
38   // PowerPC64 ELF.
39   if (Obj->getArch() == Triple::ppc64) {
40     for (section_iterator Section : Obj->sections()) {
41       StringRef Name;
42       StringRef Data;
43       if (auto EC = Section->getName(Name))
44         return EC;
45       if (Name == ".opd") {
46         if (auto EC = Section->getContents(Data))
47           return EC;
48         OpdExtractor.reset(new DataExtractor(Data, Obj->isLittleEndian(),
49                                              Obj->getBytesInAddress()));
50         OpdAddress = Section->getAddress();
51         break;
52       }
53     }
54   }
55   std::vector<std::pair<SymbolRef, uint64_t>> Symbols =
56       computeSymbolSizes(*Obj);
57   for (auto &P : Symbols)
58     res->addSymbol(P.first, P.second, OpdExtractor.get(), OpdAddress);
59 
60   // If this is a COFF object and we didn't find any symbols, try the export
61   // table.
62   if (Symbols.empty()) {
63     if (auto *CoffObj = dyn_cast<COFFObjectFile>(Obj))
64       if (auto EC = res->addCoffExportSymbols(CoffObj))
65         return EC;
66   }
67   return std::move(res);
68 }
69 
70 SymbolizableObjectFile::SymbolizableObjectFile(ObjectFile *Obj,
71                                                std::unique_ptr<DIContext> DICtx)
72     : Module(Obj), DebugInfoContext(std::move(DICtx)) {}
73 
74 namespace {
75 struct OffsetNamePair {
76   uint32_t Offset;
77   StringRef Name;
78   bool operator<(const OffsetNamePair &R) const {
79     return Offset < R.Offset;
80   }
81 };
82 }
83 
84 std::error_code SymbolizableObjectFile::addCoffExportSymbols(
85     const COFFObjectFile *CoffObj) {
86   // Get all export names and offsets.
87   std::vector<OffsetNamePair> ExportSyms;
88   for (const ExportDirectoryEntryRef &Ref : CoffObj->export_directories()) {
89     StringRef Name;
90     uint32_t Offset;
91     if (auto EC = Ref.getSymbolName(Name))
92       return EC;
93     if (auto EC = Ref.getExportRVA(Offset))
94       return EC;
95     ExportSyms.push_back(OffsetNamePair{Offset, Name});
96   }
97   if (ExportSyms.empty())
98     return std::error_code();
99 
100   // Sort by ascending offset.
101   array_pod_sort(ExportSyms.begin(), ExportSyms.end());
102 
103   // Approximate the symbol sizes by assuming they run to the next symbol.
104   // FIXME: This assumes all exports are functions.
105   uint64_t ImageBase = CoffObj->getImageBase();
106   for (auto I = ExportSyms.begin(), E = ExportSyms.end(); I != E; ++I) {
107     OffsetNamePair &Export = *I;
108     // FIXME: The last export has a one byte size now.
109     uint32_t NextOffset = I != E ? I->Offset : Export.Offset + 1;
110     uint64_t SymbolStart = ImageBase + Export.Offset;
111     uint64_t SymbolSize = NextOffset - Export.Offset;
112     SymbolDesc SD = {SymbolStart, SymbolSize};
113     Functions.insert(std::make_pair(SD, Export.Name));
114   }
115   return std::error_code();
116 }
117 
118 std::error_code SymbolizableObjectFile::addSymbol(const SymbolRef &Symbol,
119                                                   uint64_t SymbolSize,
120                                                   DataExtractor *OpdExtractor,
121                                                   uint64_t OpdAddress) {
122   Expected<SymbolRef::Type> SymbolTypeOrErr = Symbol.getType();
123   if (!SymbolTypeOrErr)
124     return errorToErrorCode(SymbolTypeOrErr.takeError());
125   SymbolRef::Type SymbolType = *SymbolTypeOrErr;
126   if (SymbolType != SymbolRef::ST_Function && SymbolType != SymbolRef::ST_Data)
127     return std::error_code();
128   ErrorOr<uint64_t> SymbolAddressOrErr = Symbol.getAddress();
129   if (auto EC = SymbolAddressOrErr.getError())
130     return EC;
131   uint64_t SymbolAddress = *SymbolAddressOrErr;
132   if (OpdExtractor) {
133     // For big-endian PowerPC64 ELF, symbols in the .opd section refer to
134     // function descriptors. The first word of the descriptor is a pointer to
135     // the function's code.
136     // For the purposes of symbolization, pretend the symbol's address is that
137     // of the function's code, not the descriptor.
138     uint64_t OpdOffset = SymbolAddress - OpdAddress;
139     uint32_t OpdOffset32 = OpdOffset;
140     if (OpdOffset == OpdOffset32 &&
141         OpdExtractor->isValidOffsetForAddress(OpdOffset32))
142       SymbolAddress = OpdExtractor->getAddress(&OpdOffset32);
143   }
144   Expected<StringRef> SymbolNameOrErr = Symbol.getName();
145   if (!SymbolNameOrErr)
146     return errorToErrorCode(SymbolNameOrErr.takeError());
147   StringRef SymbolName = *SymbolNameOrErr;
148   // Mach-O symbol table names have leading underscore, skip it.
149   if (Module->isMachO() && SymbolName.size() > 0 && SymbolName[0] == '_')
150     SymbolName = SymbolName.drop_front();
151   // FIXME: If a function has alias, there are two entries in symbol table
152   // with same address size. Make sure we choose the correct one.
153   auto &M = SymbolType == SymbolRef::ST_Function ? Functions : Objects;
154   SymbolDesc SD = { SymbolAddress, SymbolSize };
155   M.insert(std::make_pair(SD, SymbolName));
156   return std::error_code();
157 }
158 
159 // Return true if this is a 32-bit x86 PE COFF module.
160 bool SymbolizableObjectFile::isWin32Module() const {
161   auto *CoffObject = dyn_cast<COFFObjectFile>(Module);
162   return CoffObject && CoffObject->getMachine() == COFF::IMAGE_FILE_MACHINE_I386;
163 }
164 
165 uint64_t SymbolizableObjectFile::getModulePreferredBase() const {
166   if (auto *CoffObject = dyn_cast<COFFObjectFile>(Module))
167     return CoffObject->getImageBase();
168   return 0;
169 }
170 
171 bool SymbolizableObjectFile::getNameFromSymbolTable(SymbolRef::Type Type,
172                                                     uint64_t Address,
173                                                     std::string &Name,
174                                                     uint64_t &Addr,
175                                                     uint64_t &Size) const {
176   const auto &SymbolMap = Type == SymbolRef::ST_Function ? Functions : Objects;
177   if (SymbolMap.empty())
178     return false;
179   SymbolDesc SD = { Address, Address };
180   auto SymbolIterator = SymbolMap.upper_bound(SD);
181   if (SymbolIterator == SymbolMap.begin())
182     return false;
183   --SymbolIterator;
184   if (SymbolIterator->first.Size != 0 &&
185       SymbolIterator->first.Addr + SymbolIterator->first.Size <= Address)
186     return false;
187   Name = SymbolIterator->second.str();
188   Addr = SymbolIterator->first.Addr;
189   Size = SymbolIterator->first.Size;
190   return true;
191 }
192 
193 bool SymbolizableObjectFile::shouldOverrideWithSymbolTable(
194     FunctionNameKind FNKind, bool UseSymbolTable) const {
195   // When DWARF is used with -gline-tables-only / -gmlt, the symbol table gives
196   // better answers for linkage names than the DIContext. Otherwise, we are
197   // probably using PEs and PDBs, and we shouldn't do the override. PE files
198   // generally only contain the names of exported symbols.
199   return FNKind == FunctionNameKind::LinkageName && UseSymbolTable &&
200          isa<DWARFContext>(DebugInfoContext.get());
201 }
202 
203 DILineInfo SymbolizableObjectFile::symbolizeCode(uint64_t ModuleOffset,
204                                                  FunctionNameKind FNKind,
205                                                  bool UseSymbolTable) const {
206   DILineInfo LineInfo;
207   if (DebugInfoContext) {
208     LineInfo = DebugInfoContext->getLineInfoForAddress(
209         ModuleOffset, getDILineInfoSpecifier(FNKind));
210   }
211   // Override function name from symbol table if necessary.
212   if (shouldOverrideWithSymbolTable(FNKind, UseSymbolTable)) {
213     std::string FunctionName;
214     uint64_t Start, Size;
215     if (getNameFromSymbolTable(SymbolRef::ST_Function, ModuleOffset,
216                                FunctionName, Start, Size)) {
217       LineInfo.FunctionName = FunctionName;
218     }
219   }
220   return LineInfo;
221 }
222 
223 DIInliningInfo SymbolizableObjectFile::symbolizeInlinedCode(
224     uint64_t ModuleOffset, FunctionNameKind FNKind, bool UseSymbolTable) const {
225   DIInliningInfo InlinedContext;
226 
227   if (DebugInfoContext)
228     InlinedContext = DebugInfoContext->getInliningInfoForAddress(
229         ModuleOffset, getDILineInfoSpecifier(FNKind));
230   // Make sure there is at least one frame in context.
231   if (InlinedContext.getNumberOfFrames() == 0)
232     InlinedContext.addFrame(DILineInfo());
233 
234   // Override the function name in lower frame with name from symbol table.
235   if (shouldOverrideWithSymbolTable(FNKind, UseSymbolTable)) {
236     std::string FunctionName;
237     uint64_t Start, Size;
238     if (getNameFromSymbolTable(SymbolRef::ST_Function, ModuleOffset,
239                                FunctionName, Start, Size)) {
240       InlinedContext.getMutableFrame(InlinedContext.getNumberOfFrames() - 1)
241           ->FunctionName = FunctionName;
242     }
243   }
244 
245   return InlinedContext;
246 }
247 
248 DIGlobal SymbolizableObjectFile::symbolizeData(uint64_t ModuleOffset) const {
249   DIGlobal Res;
250   getNameFromSymbolTable(SymbolRef::ST_Data, ModuleOffset, Res.Name, Res.Start,
251                          Res.Size);
252   return Res;
253 }
254 
255 }  // namespace symbolize
256 }  // namespace llvm
257 
258